1 / 23

Advanced Java Programming

Advanced Java Programming. Umar Kalim Dept. of Communication Systems Engineering umar.kalim@niit.edu.pk http://www.niit.edu.pk/~umarkalim 29/01/2008. Agenda. Implementation & Performance Source Stanford University ~ Lecture by Manu Kumar. Java Implementation and Performance.

zorana
Download Presentation

Advanced Java Programming

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Advanced Java Programming Umar Kalim Dept. of Communication Systems Engineering umar.kalim@niit.edu.pk http://www.niit.edu.pk/~umarkalim 29/01/2008 cs420

  2. Agenda • Implementation & Performance • Source • Stanford University ~ Lecture by Manu Kumar cs420

  3. Java Implementation and Performance • Java Compiler Structure • .java files contain source code • Compiled into .class files which contain bytecode • Bytecode • A compiled class stored in a .class file or a .jar file • Represent a computation in a portable way • As a PDF is to an image • Java Virtual machine • Abstract stack machine • Bytecode is written to run on the JVM • Program that loads and runs the bytecode • Interprets the bytecode to “run” the program • Runs code with various robustness and safety checks cs420

  4. Verifier and Bytecode • Verifier • Part of the VM that checks that bytecode is well formed • Makes Java virus proof • A malicious person can write invalid bytecode, but it will be detected by the Verifier • Usually no verifier errors since the compiler produces “correct” bytecode • Still possible to write bytecode by hand • Bytecode example • javap –c will print the actual bytecode for a class cs420

  5. JITs and Hotspot • Just-In-Time Compiler • JVM may compile the bytecode into native code at runtime • Maintains robustness/safety checks (slow startup) • HotSpot • Does a sophisticated runtime optimization for which part to compile • Sometimes does a better job than native C++ code since it has more information about the running program • Future • May cache the compiled version to speed up class loading • Bytecode is a distribution format • Similar to PDF cs420

  6. Optimization Quotes • Rules of Optimization • Rule 1: Don’t do it. • Rule 2 (for experts only): Don’t do it yet. • M.A. Jackson • “More computing sins are committed in the name of efficiency (without necessarily achieving it) that for any other reason – including blind stupidity.” – W.A. Wulf • Y2K bug! – saving 2 bytes! • We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil.” – Donald Knuth cs420

  7. Optimization ~ sunehri asool! • Reality • Hard to predict where the bottlenecks are • Easier to use tools to measure the bottlenecks once the code is written • Write the code you want to be correct and finished first, then worry about optimization • “Premature Optimization” = evil • Classic advice from Don Knuth • Write the code to be straightforward and correct first • May already be fast enough! • If not, measure the bottleneck • Focus optimization on bottleneck using Algorithms and Language optimizations cs420

  8. Optimization ~ sunehri asool! • Data Structures • Have a profound influence on performance • Early design helps once • Choice of data structure can constrain what algorithms you can use • Proportionality to Caller • Foo() takes 1 ms. Bar() calls foo. • If Bar() takes 20 ms, it’s not worth looking at Foo() • If Bar() takes 2ms, then we should look at Foo() cs420

  9. Optimization ~ sunehri asool! • 1-1 User Event Rule • If something happened a fixed number of times (1-3) for each user event, then it’s not worth looking at… • If something happens 100s of times for each user it is worth looking at…? • User events are very slow from the CPU’s perspective cs420

  10. Optimization ~ sunehri asool! Good Architect ~ Design Principles + Experience cs420

  11. Java Optimization Tip #1 • 1-10-100 Rule • Assignment – 1 unit of time • Method call – 10 units of time • New Object or Array – 100 units of time • Rule of thumb only. Not scientific. • Hard to determine the actual cost • Also sometimes known as the 1-10-1000 rule, but modern GC is much more efficient • Bad idea to try and maintain your own free list. The GC knows best. cs420

  12. Java Optimization Tip #2 • int getWidth() vs. Dimension getSize() • getSize() requires a heap allocated object • getWidth() and getHeight() may just be inlined to move the two ints right into the local variables of the caller code • With Hotspot, shortlived object (like Dimension) are less of a concern… cs420

  13. Java Optimization Tip #3 • Locals are faster than Instance variables • Local (stack) variables faster than any member variables of objects • Easier for the optimizer to work with • Inside loops, pull needed values into local variables • 1. Slow: message send • … i < piece.getWidth() • 2. Medium: instance variable • … i < piece.width • 3. Fast: local variable • … final int width = piece.getWidth • … i < width • This is faster since the JIT can put the value in a native register cs420

  14. Java Optimization Tip #4 • Avoid Synchronized (Vector) • Synchronized methods have a cost associated with them • This is significantly improved in Java 1.3 • Can have synchronized and unsynchronized methods and switch based on some flag • Use “immutable” objects to finesse synchronization problems • Vector class is sychronized for everything • Use ArrayList instead! • If you can use a regular array, even better cs420

  15. Java Optimization Tip #5 • StringBuffer • Use StringBuffer for multiple append operations • Convert to String only when done • Automatic case • Compiler optimizes the case of several string +’d together on one line • String s = “a string” + foo.toString() + “more” • No: String record; // ivar void transaction(String id) { record = record + " " + id; // NO, chews through memory } • Yes: StringBuffer record; void transaction(String id) { record.append(" "); record.append(id); + id; } cs420

  16. Java Optimization Tip #6 • Don’t Parse • Obvious but slow strategy – read in XML, ASCII, etc. • Build a big data structure • Faster approach • Read into memory, but keep as characters • Search/Parse when needed • Or Parse only subparts cs420

  17. Java Optimization Tip #7 • Avoid weird code • JVM will optimize most standard coding styles • So write code in the most obvious, common way • Weird code is often the result of an attempt at optimization! • Let the JIT/Hotspot do it’s thing! cs420

  18. Java Optimization Tip #8 • Threading / GUI Threading • Use separate thread to ensure the GUI is snappy • Pros • Makes best use of parallel hardware • Cons • Software is harder to write • Bugs can be subtle • Locking/Unlocking costs cs420

  19. Java Optimization Tip #9 • Inlining Methods/Classes • JVM optimizers and HotSpot use aggressive inlining • Pasting called code into the caller code • final keyword • for a class means it will not be subclassed • for a method means it will not be overridden • Use final keyword to help the optimizer do more inlining cs420

  20. Java Optimization Tip #9 cont. Inlined Non-Inlined cs420

  21. Java Optimization Tip #9 cont. • Advantages of inlining • Values are passed from A() to B() to C() • After inlining, the values can just stay in registers • Reduced number of load/saves • Propagation of analysis • Having code inlined can often lead to better optimizations since the compiler can see values from start to finish cs420

  22. Java Optimization Tip #10 • Think about memory traffic • Old: CPU bound • New: Memory bound • CPU operations are cheaper and faster • Once data is in the cache it is cheaper to work with • Reduce the roundtrips to memory, disk, network • Linked List vs. Chunked List • Linked List: Read a node, then fetch the next node • Chunked List: Each element contains a small array of elements • Makes better use of cache lines/memory pages • … some of this is very low level!  cs420

  23. Summary & Questions? That’s all for today! cs420

More Related